Yash Ganatra

Yash Ganatra | AI/ML Engineer & Backend-Focused Full-Stack Developer

I am a Computer Science & Data Science undergraduate with a strong focus on AI engineering, Retrieval-Augmented Generation (RAG) systems, multi-agent workflows, and scalable backend development. I enjoy building production-grade AI systems that solve real business problems through clean architecture, measurable performance improvements, and reliable deployment.

About Me

Runner-up at Datathon 2025, where I built an AI-based OTT churn prediction system for customer risk segmentation and retention analysis.

Track Winner at Quasar 3.0 (ML Hackathon) for developing an NLP-driven MCQ generation system using modern language models.

Vice Chairperson (Technical) at DJS–S4DS, where I led the organization of DataHack 3.0 with over 1500 participants and engineered the end-to-end project submission portal.

Hands-on experience building RAG pipelines, multi-agent AI systems, and high-performance APIs used in real production environments.

Featured Project
FinKar – Multi-Agent AI Financial Assistant

Designed and implemented a multi-agent AI assistant using LangGraph, Pinecone-based RAG, and Groq-hosted LLMs, improving financial query accuracy by approximately 65%.

Built a high-performance FastAPI backend with optimized vector retrieval and agent orchestration, enabling nearly 60% faster data access for financial analytics.

Delivered a cross-platform React + Ionic application, optimizing rendering and animations to achieve around 50% lower UI latency and smoother real-time interactions.

Internship Experience
AI Intern – AppInSource Technologies

Architected a multi-engine RAG system using FastAPI with three isolated vector collections for regulatory content, business logic, and legacy code, reducing cross-domain retrieval noise by approximately 35%.

Implemented domain-aware chunking (450–600 tokens), metadata filtering, and re-ranking, improving top-3 retrieval precision by ~30% for KYC and onboarding workflows.

Integrated Groq-hosted LLaMA 3.x as the reasoning layer and built a React-based interface with dynamic process and architecture visualizations, reducing onboarding and internal support effort by ~40%.

Backend Developer Intern – Taag.one

Built a rule-based brand–creator recommendation engine using weighted scoring and constraint filters, improving match relevance by ~25% and reducing manual shortlisting by ~40%.

Developed modular RESTful APIs for campaign management, invoicing, and agency workflows with role-based access control.

Delivered stateless, scalable APIs with pagination and versioning to ensure reliable cross-platform usage across mobile and web clients.

AI & Machine Learning Intern – Unfluke

Fine-tuned Mistral 7B and optimized inference pipelines for text and image generation, reducing manual content creation effort by ~80%.

Designed a Python-based agent-based market simulation framework modeling heterogeneous trading agents and emergent market dynamics.

Deployed internal tools and public-facing platforms to support product demos, investor communication, and analytics-driven reporting.

Technical Skills

Programming Languages

Python, C, C++, Java, SQL, JavaScript, HTML, CSS

AI / ML & LLM Frameworks

PyTorch, TensorFlow, Scikit-learn, Hugging Face Transformers, LangChain, Ollama

Web & Backend Technologies

FastAPI, Flask, Node.js, Express, React, Ionic (Capacitor)

Databases

MySQL, SQLite, MongoDB, Neo4j

Tools & Platforms

Git, Streamlit, Hugging Face, Postman, Tableau

Achievements

Runner-up, Datathon 2025 — AI-based OTT churn prediction system

Track Winner, Quasar 3.0 — NLP-driven MCQ generation platform

Vice Chairperson (Technical), DJS–S4DS — Led DataHack 3.0 with 1500+ participants

Built and deployed production-grade RAG and multi-agent AI systems